diff --git a/benchmarks/single_node/glm5_fp8_b300.sh b/benchmarks/single_node/glm5_fp8_b300.sh
index 730cc39505..942f5acfcd 100644
--- a/benchmarks/single_node/glm5_fp8_b300.sh
+++ b/benchmarks/single_node/glm5_fp8_b300.sh
@@ -23,13 +23,17 @@ nvidia-smi
 
 if [[ "$MODEL" != /* ]]; then hf download "$MODEL"; fi
 
-pip install --no-deps "transformers==5.2.0" "huggingface-hub==1.4.1"
-
-# Workaround for sgl-project/sglang#25551: v0.5.12 DeepGemm TMA-descriptor
-# regression on B300 (sm_120) crashes CUDA graph capture with
-# CUDA_ERROR_ILLEGAL_ADDRESS. Disabling JIT DeepGemm bypasses the affected
-# kernel path. Restore to =1 once the upstream regression is fixed.
-export SGL_ENABLE_JIT_DEEPGEMM=0
+pip install --break-system-packages --no-deps "transformers==5.2.0" "huggingface-hub==1.4.1"
+
+# Testing @trevor-m's suggestion in sgl-project/sglang#25551 (comment 4481466979):
+# downgrade sgl-deep-gemm 0.1.0 → 0.0.1 inside the v0.5.12 container to check
+# whether the deepgemm version jump is what causes the B300 TMA-descriptor
+# CUDA_ERROR_ILLEGAL_ADDRESS regression. Re-enabling JIT DeepGemm so the
+# downgraded version actually runs.
+# --break-system-packages required: the container's Python is PEP-668 externally-managed,
+# so the previous attempt silently failed and left the bundled 0.1.0 in place.
+pip install --break-system-packages --no-deps "sgl-deep-gemm==0.0.1"
+export SGL_ENABLE_JIT_DEEPGEMM=1
 
 SERVER_LOG=/workspace/server.log
 PORT=${PORT:-8888}
diff --git a/benchmarks/single_node/glm5_fp8_b300_mtp.sh b/benchmarks/single_node/glm5_fp8_b300_mtp.sh
index 0d4290dd32..d8b6d96adc 100755
--- a/benchmarks/single_node/glm5_fp8_b300_mtp.sh
+++ b/benchmarks/single_node/glm5_fp8_b300_mtp.sh
@@ -23,13 +23,17 @@ nvidia-smi
 
 if [[ "$MODEL" != /* ]]; then hf download "$MODEL"; fi
 
-pip install --no-deps "transformers==5.2.0" "huggingface-hub==1.4.1"
-
-# Workaround for sgl-project/sglang#25551: v0.5.12 DeepGemm TMA-descriptor
-# regression on B300 (sm_120) crashes CUDA graph capture with
-# CUDA_ERROR_ILLEGAL_ADDRESS. Disabling JIT DeepGemm bypasses the affected
-# kernel path. Restore to =1 once the upstream regression is fixed.
-export SGL_ENABLE_JIT_DEEPGEMM=0
+pip install --break-system-packages --no-deps "transformers==5.2.0" "huggingface-hub==1.4.1"
+
+# Testing @trevor-m's suggestion in sgl-project/sglang#25551 (comment 4481466979):
+# downgrade sgl-deep-gemm 0.1.0 → 0.0.1 inside the v0.5.12 container to check
+# whether the deepgemm version jump is what causes the B300 TMA-descriptor
+# CUDA_ERROR_ILLEGAL_ADDRESS regression. Re-enabling JIT DeepGemm so the
+# downgraded version actually runs.
+# --break-system-packages required: the container's Python is PEP-668 externally-managed,
+# so the previous attempt silently failed and left the bundled 0.1.0 in place.
+pip install --break-system-packages --no-deps "sgl-deep-gemm==0.0.1"
+export SGL_ENABLE_JIT_DEEPGEMM=1
 export SGLANG_ENABLE_SPEC_V2=1
 
 SERVER_LOG=/workspace/server.log
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index c62d947816..90387fd42b 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -3043,3 +3043,10 @@
   description:
     - "Update SGLang image from v0.5.11-cu130 (5d old) to v0.5.12-cu130"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1475
+
+- config-keys:
+    - glm5-fp8-b300-sglang
+    - glm5-fp8-b300-sglang-mtp
+  description:
+    - "Test @trevor-m's suggestion in sgl-project/sglang#25551: pin sgl-deep-gemm==0.0.1 inside v0.5.12 container to isolate whether the deep-gemm 0.0.1→0.1.0 upgrade is the source of the B300 CUDA_ERROR_ILLEGAL_ADDRESS regression."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1512